統計学輪講(第24回) 日時 2017年12月05日(火) 14時55分~15時45分 場所 経済学部新棟3階第3教室 講演者 粟屋 直 (経済D1) 演題 Particle rolling MCMC with forward and backward block sampling with application to stochastic volatility models 概要 The objective is to provide a new simulation-based methodology for rolling estimation in state space model from Bayesian approach. This type of estimation requires sampling by simulation-based method from a lot of posteriors if the model does not have so simple form. Repetition of sampling from each posterior by Markov Chain Monte Carlo is not realistic from a viewpoint of computational time, so in order to address this problem a new sampling algorithm based on sequential Monte Carlo is presented. This method is applied to SP 500 data with the realized stochastic volatility with leverage model and how the economic structure which generates the financial data is changed is shown.